Module Records Blog has a nice report on options for hardware speech synthesis (this is following on from the singing robots post). For $150 you can buy a complete TextSpeak module, which takes the output from a PS2 keyboard and turns it into speech. $1,995 (!) gets a VoiceNote mPower QT, which does the same sort of thing, but presumably much better. But the cool option is Speakjet - a one-chip speech synth. Just $55 gets you the Speakjet development kit (including the chip). It's a circuit board with switches and LEDs and a battery and a speaker output, plus various analog and digital ins and outs. Stanley from Module Records has one and says: "The chip sounds very good with delicious robotic feeling. I plan to make a computer controlled musical speech synthesizer with it, so stay tuned..." He won't be the first: Here [mp3] is a demo of a SpeakJet chip being controlled by a modular synth (using a PSIM module).
Posted by Tom Whitwell.
Comments:
DAMMIT! Someone's gonna beat me to the punch. I've been working on sing speech synths (concatonated), as well. However, I'm sure no matter who starts it, "it" will be a trend. Soon, everyone will have robots singing that sound almost human. And everyone will get sick of it.
Try this.... Make a big multisample of yourself saying all the phonetics. ("Ah, buh cuh deh eff fu gu hu ih juh kuh luh muh nuh" etc, "ay bee cee dee, ee, eff" etc). And some sss, khh sounds. Make it so you have each sample on it's own key. Then play sentences with your own voice! Kept me amused for hours. :) Making it monophonic and adding an occasional pitch bend helps it sound real in an unnatural kind of way.
I actually dreamed a while back that I found a hardware rackmount speech synth at a car boot sale, and made a 7" of just synth singing, and John Peel played it... ;( Alas the dream didn't come true...
"Make a big multisample of yourself saying all the phonetics."
I've gone so far as to record all the vowels and consonants, AND add loop points to vowel sounds I've sung. I then created a Soundfont out of it. The first time I did it, the trouble was the transitions from vowel to soft consonants ("oo" to "mmm"). So now, I'm going to start recording again, but this time record every consonant-vowel and vowel-consonant combination possible.
"When you've finished, I trust you'll mail me the finished soundfont, so it can be shared with the MT reading public?"
I sure will. Unfortunately, the combinations I've come up with is around 600, way too many for a soundfont to lay out on a 9-octave keyboard. I may have to get SynthEdit involved and create a VSTi.
Ok well, speech synths are pretty cool, but it seems like they sounded "cooler" back in the late 70's ala Kwerk "radioactivity". What the hell were Kraftwerk using back then?
The newish ones just don't seem to have the gutteral dynamism of the old ones, sorta like virtual analog vs real analog.
And no more bent Speak and Spells! That train has left the station.
Kraftwerk where using a TexasInstruments "Language Translator" for the robot voices. More info on the hardware can be found here http://www.datamath.org/ in the "speech products" section